[Mlir-commits] [mlir] [mlir][transform] Add support for transform.param pad multiples in `PadOp` (PR #90755)
llvmlistbot at llvm.org
llvmlistbot at llvm.org
Wed May 1 12:03:31 PDT 2024
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-mlir-linalg
Author: None (srcarroll)
<details>
<summary>Changes</summary>
This patch modifies the definition of `PadOp` to take transform params and handles for the `pad_to_multiple_of` operand.
---
Full diff: https://github.com/llvm/llvm-project/pull/90755.diff
5 Files Affected:
- (modified) mlir/include/mlir/Dialect/Linalg/TransformOps/LinalgTransformOps.td (+19-11)
- (modified) mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp (+161-7)
- (modified) mlir/python/mlir/dialects/transform/structured.py (+15-2)
- (modified) mlir/test/Dialect/Linalg/transform-op-pad.mlir (+37-2)
- (modified) mlir/test/python/dialects/transform_structured_ext.py (+3-2)
``````````diff
diff --git a/mlir/include/mlir/Dialect/Linalg/TransformOps/LinalgTransformOps.td b/mlir/include/mlir/Dialect/Linalg/TransformOps/LinalgTransformOps.td
index d0ad4ccdf031d9..ada7f7666d5f60 100644
--- a/mlir/include/mlir/Dialect/Linalg/TransformOps/LinalgTransformOps.td
+++ b/mlir/include/mlir/Dialect/Linalg/TransformOps/LinalgTransformOps.td
@@ -978,8 +978,8 @@ def PackTransposeOp : Op<Transform_Dialect, "structured.pack_transpose", [
//===----------------------------------------------------------------------===//
def PadOp : Op<Transform_Dialect, "structured.pad",
- [FunctionalStyleTransformOpTrait, MemoryEffectsOpInterface,
- DeclareOpInterfaceMethods<TransformOpInterface>,
+ [FunctionalStyleTransformOpTrait, DeclareOpInterfaceMethods<MemoryEffectsOpInterface>,
+ TransformOpInterface,
ReportTrackingListenerFailuresOpTrait]> {
let description = [{
Pads the operations pointed to by the target handle using the options
@@ -1011,7 +1011,9 @@ def PadOp : Op<Transform_Dialect, "structured.pad",
(ins TransformHandleTypeInterface:$target,
DefaultValuedAttr<ArrayAttr, "{}">:$padding_values,
DefaultValuedAttr<I64ArrayAttr, "{}">:$padding_dimensions,
- OptionalAttr<I64ArrayAttr>:$pad_to_multiple_of,
+ Variadic<TransformAnyParamTypeOrAnyHandle>:$pad_to_multiple_of,
+ DefaultValuedOptionalAttr<DenseI64ArrayAttr, "{}">:
+ $static_pad_to_multiple_of,
DefaultValuedAttr<I64ArrayAttr, "{}">:$pack_paddings,
DefaultValuedAttr<
TypedArrayAttrBase<I64ArrayAttr, "array of arrays of i64">,
@@ -1021,8 +1023,7 @@ def PadOp : Op<Transform_Dialect, "structured.pad",
TransformHandleTypeInterface:$pad,
TransformHandleTypeInterface:$copy);
- let assemblyFormat =
- "$target attr-dict `:` functional-type(operands, results)";
+ let hasCustomAssemblyFormat = 1;
let hasVerifier = 1;
let builders = [
@@ -1033,7 +1034,13 @@ def PadOp : Op<Transform_Dialect, "structured.pad",
// TODO: support other operations (e.g. min, max etc).
OpBuilder<(ins "Value":$target,
"ArrayRef<int64_t>":$paddingDimensions,
- CArg<"ArrayRef<int64_t>", "{}">:$padToMultipleOf,
+ CArg<"ArrayRef<int64_t>", "{}">:$staticPadToMultipleOf,
+ CArg<"ArrayRef<int64_t>", "{}">:$packPaddings,
+ CArg<"ArrayRef<Attribute>", "{}">:$transposePaddings,
+ CArg<"StringRef", "::mlir::bufferization::MaterializeInDestinationOp::getOperationName()">:$copyBackOp)>,
+ OpBuilder<(ins "Value":$target,
+ "ArrayRef<int64_t>":$paddingDimensions,
+ "ArrayRef<OpFoldResult>":$mixedPadToMultipleOf,
CArg<"ArrayRef<int64_t>", "{}">:$packPaddings,
CArg<"ArrayRef<Attribute>", "{}">:$transposePaddings,
CArg<"StringRef", "::mlir::bufferization::MaterializeInDestinationOp::getOperationName()">:$copyBackOp)>
@@ -1043,11 +1050,12 @@ def PadOp : Op<Transform_Dialect, "structured.pad",
/// copy_back_op attribute value indicating that no copy back is desired.
static constexpr StringRef kCopyOpNone = "none";
- ::mlir::DiagnosedSilenceableFailure applyToOne(
- ::mlir::transform::TransformRewriter &rewriter,
- ::mlir::linalg::LinalgOp target,
- ::mlir::transform::ApplyToEachResultList &results,
- ::mlir::transform::TransformState &state);
+ SmallVector<OpFoldResult> getMixedPadToMultipleOf();
+
+ ::mlir::DiagnosedSilenceableFailure apply(
+ ::mlir::transform::TransformRewriter &rewriter,
+ ::mlir::transform::TransformResults &results,
+ ::mlir::transform::TransformState &state);
}];
}
diff --git a/mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp b/mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp
index 156784f0e67402..c68bc4c1f025ac 100644
--- a/mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp
+++ b/mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp
@@ -1664,6 +1664,8 @@ transform::PackTransposeOp::apply(transform::TransformRewriter &rewriter,
// PadOp
//===---------------------------------------------------------------------===//
+static const StringLiteral kPadToMultipleOfKeyword = "pad_to_multiple_of";
+
void transform::PadOp::build(OpBuilder &b, OperationState &result, Value target,
ArrayRef<int64_t> paddingDimensions,
ArrayRef<int64_t> padToMultipleOf,
@@ -1677,14 +1679,121 @@ void transform::PadOp::build(OpBuilder &b, OperationState &result, Value target,
/*target=*/target,
/*paddingValues=*/ArrayAttr(), // let inference handle this
/*paddingDimensions=*/b.getI64ArrayAttr(paddingDimensions),
+ /*padToMultipleOf=*/ValueRange{},
/*padToMultipleOf=*/
- (padToMultipleOf.empty() ? ArrayAttr()
- : b.getI64ArrayAttr(padToMultipleOf)),
+ (padToMultipleOf.empty()
+ ? DenseI64ArrayAttr()
+ : b.getDenseI64ArrayAttr(padToMultipleOf)),
+ /*packPaddings=*/b.getI64ArrayAttr(packPaddings),
+ /*transposePaddings=*/b.getArrayAttr(transposePaddings),
+ /*copyBackOp=*/b.getStringAttr(copyBackOp));
+}
+
+void transform::PadOp::build(OpBuilder &b, OperationState &result, Value target,
+ ArrayRef<int64_t> paddingDimensions,
+ ArrayRef<OpFoldResult> mixedPadToMultipleOf,
+ ArrayRef<int64_t> packPaddings,
+ ArrayRef<Attribute> transposePaddings,
+ StringRef copyBackOp) {
+ auto resultType = transform::AnyOpType::get(b.getContext());
+ SmallVector<int64_t> staticPadToMultipleOf;
+ SmallVector<Value> dynamicPadToMultipleOf;
+ dispatchIndexOpFoldResults(mixedPadToMultipleOf, dynamicPadToMultipleOf,
+ staticPadToMultipleOf);
+ return build(/*builder=*/b,
+ /*result=*/result,
+ /*types=*/TypeRange{resultType, resultType},
+ /*target=*/target,
+ /*paddingValues=*/ArrayAttr(), // let inference handle this
+ /*paddingDimensions=*/b.getI64ArrayAttr(paddingDimensions),
+ /*padToMultipleOf=*/dynamicPadToMultipleOf,
+ /*padToMultipleOf=*/staticPadToMultipleOf,
/*packPaddings=*/b.getI64ArrayAttr(packPaddings),
/*transposePaddings=*/b.getArrayAttr(transposePaddings),
/*copyBackOp=*/b.getStringAttr(copyBackOp));
}
+void PadOp::getEffects(
+ SmallVectorImpl<MemoryEffects::EffectInstance> &effects) {
+ consumesHandle(getTarget(), effects);
+ onlyReadsHandle(getPadToMultipleOf(), effects);
+ producesHandle(getPadded(), effects);
+ producesHandle(getPad(), effects);
+ producesHandle(getCopy(), effects);
+ modifiesPayload(effects);
+}
+
+SmallVector<OpFoldResult> PadOp::getMixedPadToMultipleOf() {
+ OpBuilder b(getContext());
+ return getMixedValues(getStaticPadToMultipleOf(), getPadToMultipleOf(), b);
+}
+
+ParseResult transform::PadOp::parse(OpAsmParser &parser,
+ OperationState &result) {
+ OpAsmParser::UnresolvedOperand target;
+ SmallVector<OpAsmParser::UnresolvedOperand> dynamicPadToMultipleOf;
+ DenseI64ArrayAttr padToMultipleOf;
+ FunctionType functionalType;
+ llvm::SMLoc operandLoc;
+
+ if (parser.parseOperand(target) || parser.getCurrentLocation(&operandLoc))
+ return ParseResult::failure();
+
+ if (succeeded(parser.parseOptionalKeyword(kPadToMultipleOfKeyword))) {
+ if (failed(parseDynamicIndexList(parser, dynamicPadToMultipleOf,
+ padToMultipleOf)))
+ return ParseResult::failure();
+ }
+
+ if (parser.parseOptionalAttrDict(result.attributes) ||
+ parser.parseColonType(functionalType) ||
+ parser.resolveOperand(target, functionalType.getInputs().front(),
+ result.operands) ||
+ parser.resolveOperands(dynamicPadToMultipleOf,
+ functionalType.getInputs().drop_front(),
+ operandLoc, result.operands))
+ return ParseResult::failure();
+
+ if (padToMultipleOf)
+ result.addAttribute(getStaticPadToMultipleOfAttrName(result.name),
+ padToMultipleOf);
+
+ result.addTypes(functionalType.getResults());
+
+ return success();
+}
+
+void transform::PadOp::print(OpAsmPrinter &p) {
+ p << ' ' << getTarget() << ' ';
+ if (!getMixedPadToMultipleOf().empty()) {
+ p << kPadToMultipleOfKeyword << ' ';
+ printDynamicIndexList(p, getOperation(), getPadToMultipleOf(),
+ getStaticPadToMultipleOfAttr(),
+ /*valueTypes=*/{},
+ /*scalables=*/{}, OpAsmParser::Delimiter::Square);
+ }
+
+ OpBuilder builder((*this)->getContext());
+ SmallVector<StringRef, 6> elidedAttrs({getStaticPadToMultipleOfAttrName()});
+ if (getCopyBackOpAttr() ==
+ builder.getStringAttr(
+ bufferization::MaterializeInDestinationOp::getOperationName()))
+ elidedAttrs.push_back(getCopyBackOpAttrName());
+ if (getPackPaddingsAttr() == builder.getI64ArrayAttr({}))
+ elidedAttrs.push_back(getPackPaddingsAttrName());
+ if (getTransposePaddingsAttr() == builder.getI64ArrayAttr({}))
+ elidedAttrs.push_back(getTransposePaddingsAttrName());
+ if (getPaddingDimensionsAttr() == builder.getI64ArrayAttr({}))
+ elidedAttrs.push_back(getPaddingDimensionsAttrName());
+ if (getPaddingValuesAttr() == builder.getArrayAttr({}))
+ elidedAttrs.push_back(getPaddingValuesAttrName());
+
+ p.printOptionalAttrDict((*this)->getAttrs(),
+ /*elidedAttrs=*/elidedAttrs);
+ p << " : ";
+ p.printFunctionalType(getOperands().getTypes(), getResults().getTypes());
+}
+
DiagnosedSilenceableFailure
transform::PadOp::apply(transform::TransformRewriter &rewriter,
transform::TransformResults &results,
@@ -1749,10 +1858,55 @@ transform::PadOp::apply(transform::TransformRewriter &rewriter,
LinalgPaddingOptions options;
options.paddingDimensions =
extractFromIntegerArrayAttr<int64_t>(getPaddingDimensions());
- SmallVector<int64_t> padToMultipleOf(options.paddingDimensions.size(), 1);
- if (getPadToMultipleOf().has_value())
+
+ SmallVector<int64_t> padToMultipleOf;
+ for (OpFoldResult sz : getMixedPadToMultipleOf()) {
+ if (sz.is<Attribute>()) {
+ auto attr = sz.get<Attribute>();
+ padToMultipleOf.push_back(cast<IntegerAttr>(attr).getInt());
+ continue;
+ } else if (sz.is<Value>() && isa<ParamType>(sz.get<Value>().getType())) {
+ ArrayRef<Attribute> params = state.getParams(sz.get<Value>());
+ if (params.size() != 1)
+ return emitSilenceableFailure(getLoc()) << "expected a single param";
+ padToMultipleOf.push_back(
+ cast<IntegerAttr>(params.front()).getValue().getSExtValue());
+ continue;
+ }
+
+ auto szPayloads = state.getPayloadOps(sz.get<Value>());
+ if (!llvm::hasSingleElement(szPayloads)) {
+ auto diag = this->emitOpError("requires pad_to_multiple_of handle that "
+ "is mapped to 1 payload op");
+ diag.attachNote(sz.get<Value>().getLoc())
+ << "mapped to " << llvm::range_size(szPayloads) << " payload ops";
+ return DiagnosedSilenceableFailure::definiteFailure();
+ }
+
+ Operation *szPayloadOp = *szPayloads.begin();
+ if (szPayloadOp->getNumResults() != 1 ||
+ !szPayloadOp->getResult(0).getType().isIndex()) {
+ auto diag = this->emitOpError(
+ "requires vector pad_to_multiple_of op with 1 index result");
+ diag.attachNote(szPayloadOp->getLoc())
+ << "pad_to_multiple_of payload op";
+ return DiagnosedSilenceableFailure::definiteFailure();
+ }
+
+ IntegerAttr attr;
+ if (!matchPattern(szPayloadOp->getResult(0), m_Constant(&attr))) {
+ auto diag = this->emitOpError("requires constant pad_to_multiple_of");
+ diag.attachNote(szPayloadOp->getLoc())
+ << "pad_to_multiple_of payload op";
+ return DiagnosedSilenceableFailure::definiteFailure();
+ }
+
+ padToMultipleOf.push_back(attr.getInt());
+ }
+ if (padToMultipleOf.empty())
padToMultipleOf =
- extractFromIntegerArrayAttr<int64_t>(*getPadToMultipleOf());
+ SmallVector<int64_t>(options.paddingDimensions.size(), 1);
+
options.padToMultipleOf = padToMultipleOf;
options.paddingValues = paddingValues;
options.packPaddings = packPaddings;
@@ -1819,8 +1973,8 @@ LogicalResult transform::PadOp::verify() {
"integers, found "
<< getPaddingDimensions();
}
- if (getPadToMultipleOf().has_value()) {
- if (getPadToMultipleOf()->size() != paddingDimensions.size()) {
+ if (!getMixedPadToMultipleOf().empty()) {
+ if (getMixedPadToMultipleOf().size() != paddingDimensions.size()) {
return emitOpError() << "expects as many multiples as padding_dimensions";
}
}
diff --git a/mlir/python/mlir/dialects/transform/structured.py b/mlir/python/mlir/dialects/transform/structured.py
index d7b41c0bd2207d..4f4a0e598df7d3 100644
--- a/mlir/python/mlir/dialects/transform/structured.py
+++ b/mlir/python/mlir/dialects/transform/structured.py
@@ -373,10 +373,11 @@ class PadOp(PadOp):
def __init__(
self,
target: Union[Operation, OpView, Value],
+ pad_to_multiple_of: Optional[Union[DynamicIndexList, ArrayAttr]] = None,
*,
padding_values: Optional[Union[ArrayAttr, Sequence[Attribute]]] = None,
padding_dimensions: OptionalIntList = None,
- pad_to_multiple_of: OptionalIntList = None,
+ static_pad_to_multiple_of: OptionalIntList = None,
pack_paddings: OptionalIntList = None,
transpose_paddings: Optional[
Union[ArrayAttr, Sequence[Union[ArrayAttr, IntOrAttrList]]]
@@ -385,6 +386,17 @@ def __init__(
loc=None,
ip=None,
):
+ if static_pad_to_multiple_of is None and pad_to_multiple_of is None:
+ dynamic_pad_to_multiple_of = []
+ elif static_pad_to_multiple_of is None:
+ (
+ dynamic_pad_to_multiple_of,
+ static_pad_to_multiple_of,
+ _,
+ ) = _dispatch_dynamic_index_list(pad_to_multiple_of)
+ else:
+ dynamic_pad_to_multiple_of = pad_to_multiple_of
+
transpose_paddings = _get_int_array_array_attr(transpose_paddings)
any_op_type = transform.AnyOpType.get()
@@ -393,9 +405,10 @@ def __init__(
any_op_type,
any_op_type,
target,
+ pad_to_multiple_of=dynamic_pad_to_multiple_of,
padding_values=padding_values,
padding_dimensions=padding_dimensions,
- pad_to_multiple_of=pad_to_multiple_of,
+ static_pad_to_multiple_of=static_pad_to_multiple_of,
pack_paddings=pack_paddings,
transpose_paddings=transpose_paddings,
copy_back_op=copy_back_op,
diff --git a/mlir/test/Dialect/Linalg/transform-op-pad.mlir b/mlir/test/Dialect/Linalg/transform-op-pad.mlir
index d27276cda49dc4..47bb5ddf4afc3e 100644
--- a/mlir/test/Dialect/Linalg/transform-op-pad.mlir
+++ b/mlir/test/Dialect/Linalg/transform-op-pad.mlir
@@ -73,10 +73,9 @@ func.func @pad_to_multiple(%arg0: tensor<24x12xf32>,
module attributes {transform.with_named_sequence} {
transform.named_sequence @__transform_main(%arg1: !transform.any_op {transform.readonly}) {
%0 = transform.structured.match ops{["linalg.matmul"]} in %arg1 : (!transform.any_op) -> !transform.any_op
- %padded, %pad, %copy_back = transform.structured.pad %0 {
+ %padded, %pad, %copy_back = transform.structured.pad %0 pad_to_multiple_of [2, 2, 1] {
padding_values=[0.0 : f32, 0.0 : f32, 0.0 : f32],
padding_dimensions=[0, 1, 2],
- pad_to_multiple_of=[2, 2, 1],
pack_paddings=[1, 1, 0]
} : (!transform.any_op) -> (!transform.any_op, !transform.any_op, !transform.any_op)
transform.yield
@@ -87,6 +86,42 @@ module attributes {transform.with_named_sequence} {
#map = affine_map<()[s0] -> (-s0 + 12, 7)>
+// CHECK-LABEL: @parametrized_pad_to_multiple
+func.func @parametrized_pad_to_multiple(%arg0: tensor<24x12xf32>,
+ %arg1: tensor<12x25xf32>,
+ %arg2: tensor<24x25xf32>,
+ %iv0 : index, %iv1 : index, %iv2 : index) -> tensor<24x25xf32> {
+ %0 = affine.min #map()[%iv2]
+ %1 = tensor.extract_slice %arg0[%iv0, %iv2] [4, %0] [1, 1] : tensor<24x12xf32> to tensor<4x?xf32>
+ %2 = tensor.extract_slice %arg1[%iv2, %iv1] [%0, 5] [1, 1] : tensor<12x25xf32> to tensor<?x5xf32>
+ %3 = tensor.extract_slice %arg2[%iv0, %iv1] [4, 5] [1, 1] : tensor<24x25xf32> to tensor<4x5xf32>
+
+ // CHECK: linalg.matmul
+ // CHECK-SAME: ins(%{{.*}}, %{{.*}} : tensor<4x7xf32>, tensor<7x6xf32>)
+ // CHECK-SAME: outs(%{{.*}} : tensor<4x6xf32>)
+ %4 = linalg.matmul ins(%1, %2 : tensor<4x?xf32>, tensor<?x5xf32>) outs(%3 : tensor<4x5xf32>) -> tensor<4x5xf32>
+ %5 = tensor.insert_slice %4 into %arg2[%iv0, %iv1] [4, 5] [1, 1] : tensor<4x5xf32> into tensor<24x25xf32>
+ func.return %5 : tensor<24x25xf32>
+}
+
+
+module attributes {transform.with_named_sequence} {
+ transform.named_sequence @__transform_main(%arg1: !transform.any_op {transform.readonly}) {
+ %0 = transform.structured.match ops{["linalg.matmul"]} in %arg1 : (!transform.any_op) -> !transform.any_op
+ %c2 = transform.param.constant 2 : i64 -> !transform.param<i64>
+ %padded, %pad, %copy_back = transform.structured.pad %0 pad_to_multiple_of [%c2, 2, 1] {
+ padding_values=[0.0 : f32, 0.0 : f32, 0.0 : f32],
+ padding_dimensions=[0, 1, 2],
+ pack_paddings=[1, 1, 0]
+ } : (!transform.any_op, !transform.param<i64>) -> (!transform.any_op, !transform.any_op, !transform.any_op)
+ transform.yield
+ }
+}
+
+// -----
+
+#map = affine_map<()[s0] -> (-s0 + 12, 7)>
+
// CHECK-LABEL: @static_sizes_output_divisible_on_empty_op
func.func @static_sizes_output_divisible_on_empty_op(%arg0: tensor<24x12xf32>,
%arg1: tensor<12x25xf32>, %arg2: tensor<24x25xf32>, %iv0: index,
diff --git a/mlir/test/python/dialects/transform_structured_ext.py b/mlir/test/python/dialects/transform_structured_ext.py
index 91ecd0fc38e174..418b1216df0532 100644
--- a/mlir/test/python/dialects/transform_structured_ext.py
+++ b/mlir/test/python/dialects/transform_structured_ext.py
@@ -315,9 +315,10 @@ def testPadOpNoArgs(target):
def testPadOpArgs(target):
structured.PadOp(
target,
+ [],
padding_values=[FloatAttr.get_f32(42.0), StringAttr.get("0")],
padding_dimensions=Attribute.parse("[1]"),
- pad_to_multiple_of=[128],
+ static_pad_to_multiple_of=[128],
pack_paddings=[0],
transpose_paddings=[[1, Attribute.parse("0")], Attribute.parse("[0, 1]")],
copy_back_op="linalg.copy",
@@ -325,9 +326,9 @@ def testPadOpArgs(target):
# CHECK-LABEL: TEST: testPadOpArgs
# CHECK: transform.sequence
# CHECK: transform.structured.pad
+ # CHECK-DAG: pad_to_multiple_of [128]
# CHECK-DAG: copy_back_op = "linalg.copy"
# CHECK-DAG: pack_paddings = [0]
- # CHECK-DAG: pad_to_multiple_of = [128]
# CHECK-DAG: padding_dimensions = [1]
# CHECK-DAG: padding_values = [4.200000e+01 : f32, "0"]
# CHECK-DAG: transpose_paddings = {{\[}}[1, 0], [0, 1]]
``````````
</details>
https://github.com/llvm/llvm-project/pull/90755
More information about the Mlir-commits
mailing list